Content
63%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is a lean, well-sequenced workflow with concrete commands and explicit filtering rules. Its main weakness is that the substantive detail — classification rules, repo policy, report template, and backtest cases — lives in six referenced files that are missing from the bundle, so the skill cannot actually be executed end-to-end as delivered.
Suggestions
Ship the six missing referenced files (repo-policy.md, relationship-rules.md, checks/fingerprint-extraction.md, checks/relationship-judgment.md, templates/report.md, validation/backtest.md) or remove them from the reference index.
Inline a minimal set of classification heuristics and the report format into SKILL.md so Steps 3 and 6 remain actionable even if detail files are absent.
Add validation checkpoints with recovery guidance, e.g., verify gh authentication and a non-empty candidate list after dedup before proceeding to classification.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is imperative and list-driven with no concept explanations, but has minor trimmable redundancy — the two relationship types are defined in the intro ("Adjacent fix… Conflict…") and restated as classification labels in Step 3, and the Composition section spends three sentences on what a sibling skill does not do — placing it at 'efficient; minor instances that could be trimmed' rather than 5. | 4 / 5 |
Actionability | Concrete commands with arguments are given for each scripted step ("scripts/extract-fingerprint.sh <pr-number>", "scripts/render-report.py < classifications.json") plus explicit caps ("top 10 by recency" per symbol), but the core classification logic is delegated to "checks/relationship-judgment.md" and the report format to "templates/report.md" — neither file exists in the bundle — so the guidance is not self-sufficient, matching 'some concrete guidance but incomplete' rather than 4's 'minor gaps'. | 3 / 5 |
Workflow Clarity | A clear six-step sequence runs from fingerprint extraction to rendered report, with concrete commands, explicit filter rules ("Remove UNRELATED and SAME_ISSUE_DIFF results"), and an evidence checkpoint ("Classify the issue as UNRELATED if this evidence is not available"). It stops short of 5 because there is no error-recovery or verification loop (e.g., what to do if a script fails or the candidate list is empty after dedup). | 4 / 5 |
Progressive Disclosure | The body is structurally well-organized with a clearly signaled, one-level-deep reference index, but scoring against the actual bundle structure, six of the seven referenced paths (repo-policy.md, relationship-rules.md, checks/fingerprint-extraction.md, checks/relationship-judgment.md, templates/report.md, validation/backtest.md) do not exist alongside the scripts — the disclosure pointers lead nowhere, matching 'references present but' organization undermined rather than 4's 'references mostly clear'. | 3 / 5 |
Total | 14 / 20 Passed |